1 From: Eric Dumazet <edumazet@google.com>
3 commit 416186fbf8c5b4e4465 ("net: Split core bits of netdev_pick_tx
4 into __netdev_pick_tx") added a bug that disables caching of queue
7 This is the source of packet reorders for TCP flows, and
8 again this is happening more often when using FQ pacing.
12 if (queue_index != old_index)
13 sk_tx_queue_set(sk, queue_index);
15 Alexander renamed the variables but forgot to change sk_tx_queue_set()
18 if (queue_index != new_index)
19 sk_tx_queue_set(sk, queue_index);
21 This means we store -1 over and over in sk->sk_tx_queue_mapping
23 Signed-off-by: Eric Dumazet <edumazet@google.com>
24 Cc: Alexander Duyck <alexander.h.duyck@intel.com>
25 Acked-by: Alexander Duyck <alexander.h.duyck@intel.com>
28 net/core/flow_dissector.c | 2 +-
29 1 file changed, 1 insertion(+), 1 deletion(-)
31 --- a/net/core/flow_dissector.c
32 +++ b/net/core/flow_dissector.c
33 @@ -347,7 +347,7 @@ u16 __netdev_pick_tx(struct net_device *
35 if (queue_index != new_index && sk &&
36 rcu_access_pointer(sk->sk_dst_cache))
37 - sk_tx_queue_set(sk, queue_index);
38 + sk_tx_queue_set(sk, new_index);
40 queue_index = new_index;